[llvm] [LV] Cost EVL width-adjusting zext/trunc as free (PR #225016)

[Pengcheng Wang via llvm-commits]

wangpc-pp wrote:

> I've also separately wondered why the return type for get.vector.length is hardcoded to i32, since the input AVL argument can be overloaded to any scalar integer type. Relaxing this to allow returning i64 would probably make life a lot easier, since there's a lot of places in the LV where we have to cast the EVL IIRC

Yeah, I have the same question when investigating the root cause. I can try to add overloaded return type for `llvm.experimental.get.vector.length` intrinsic !

https://github.com/llvm/llvm-project/pull/225016